Can Perusall Detect AI? What the Platform Sees and What It Doesn't

What Is Perusall, and Why Does the AI Question Come Up?

Perusall is a social annotation platform used alongside a course's learning management system — students read an assigned PDF, watch a video, or work through a textbook chapter, and instead of submitting a separate essay, they leave comments, questions, and replies directly on specific passages or timestamps. Other students in the same section can see and respond to those annotations, which is the core of what makes it "social" rather than a private submission. Because the graded output is a set of short annotations spread across a document rather than one continuous essay, the question of whether Perusall can detect AI comes up for a different reason than it does with Turnitin or a Canvas essay upload: the writing is shorter, more conversational, and tied to a specific quote or moment in the source material, which changes both how AI-generated annotations might look and how a scoring system would need to catch them.

Does Perusall Check for AI in Student Annotations?

No — Perusall does not run annotations through an AI-detection model, and there's no toggle in the platform for an instructor to switch one on. Perusall's own scoring engine, which the company calls Perusall's automated grading, is built to evaluate engagement signals: annotation length, whether a comment references specific content from the reading, how early the annotation was posted relative to a deadline, and whether other students engaged with it. None of those signals are designed to estimate whether the text was AI-generated. So while people sometimes ask does Perusall check for AI expecting a similarity-report-style answer, the honest answer is that the platform's algorithm was built to measure participation quality, not authorship, and it has no AI-likelihood score attached to any annotation.

Perusall's scoring engine measures whether you engaged with the reading — not whether a machine wrote your words.

What Can an Instructor Actually See on Perusall?

An instructor's view inside Perusall is more manual than automated. They can open the annotation dashboard for a class and read every comment a student left, sorted by document, student, or timestamp, and they can see the automated engagement score Perusall calculated for each student. What they cannot see, because Perusall doesn't generate it, is any kind of AI-probability percentage or highlighted "this looks machine-written" flag next to an annotation. If an instructor suspects a student's annotations were generated by AI rather than written while doing the reading, that suspicion typically comes from reading the annotations themselves — a comment that reads as generic praise of the text with no specific reference to the passage it's attached to, or phrasing that repeats across multiple readings in a way a student's own voice usually wouldn't.

Can Instructors Export Perusall Data for a Closer Review?

Yes, and this is usually the more relevant path when a professor wants to look more carefully at a student's annotation pattern. Perusall allows instructors to export annotation data, including comment text, timestamps, and engagement scores, into a spreadsheet through the platform's data export or gradebook sync features. That export doesn't run through any AI-detection model on Perusall's end — it's raw text and metadata. But once an instructor has that annotation text outside the platform, they have the option to review it manually for the same red flags they'd look for elsewhere, or to paste especially short or suspicious snippets into a separate AI detection tool if the course policy calls for that kind of check. The export step is what actually connects a Perusall assignment to any AI review — nothing automated happens before that point.

Where Do Turnitin or LMS-Level AI Checks Fit In?

Perusall is typically layered into a course through an LMS integration — most often Canvas, though also Brightspace or Blackboard — and that LMS may separately run Turnitin or another AI-detection tool on other assignments in the same course, like essays or discussion posts. That's a completely separate pipeline from Perusall's annotation grading. A course might have Turnitin's AI writing indicator active for a term paper submitted through Canvas while using Perusall purely for weekly reading annotations, and the two systems don't share data or scoring with each other. If a syllabus mentions AI detection at all, it's worth checking specifically whether that policy is describing the essay-submission tool or the reading-annotation tool, since a student could reasonably assume one covers the other when it doesn't.

How Might an Instructor Notice AI-Generated Annotations Without a Detector?

Without an automated score to lean on, instructors who suspect AI use in Perusall annotations tend to notice it through pattern-reading rather than a single flagged number.

1. Annotations that summarize a passage generically without quoting or referencing a specific sentence, word, or detail from it
2. Near-identical phrasing or sentence structure repeating across a student's comments on completely different readings
3. Annotations posted in a tight cluster right before a deadline, all similarly polished, when the reading itself was long enough to expect uneven engagement
4. Comments that never build on classmates' replies even when Perusall's threaded discussion shows other students actively responding nearby
5. A tone or vocabulary level that shifts noticeably between a student's annotations and other graded writing from the same student in the course

What Should a Student Do If Annotations Get Questioned?

If an instructor raises a concern about Perusall annotations specifically, the most useful response is usually procedural rather than defensive, since there's no Perusall-generated report to argue against in the first place.

1. Ask what specifically prompted the concern — a pattern across multiple annotations, a single comment, or something flagged after the annotations were exported and reviewed elsewhere
2. Point to genuine engagement signals Perusall already tracked, like reply threads with classmates or annotations spread naturally across the full reading rather than clustered at the end
3. If annotations were drafted with heavy help from a grammar or paraphrasing tool, explain that process honestly rather than letting the instructor assume the entire comment was AI-generated from scratch
4. Ask whether the course's AI policy actually extends to Perusall annotations or was written with essay submissions in mind, since the two are often treated differently in practice
5. If the disagreement doesn't resolve informally, ask about the course or department's formal academic integrity process rather than leaving it unaddressed

Should You Self-Check Writing Before Posting It on Perusall?

Since Perusall itself won't flag anything automatically, the practical risk sits entirely in how an annotation might read to a human instructor later, whether during a normal review or after an export. Running longer or more polished annotations through NotGPT's AI Text Detection tool before posting can highlight phrasing that reads as AI-generated, which is useful if a comment was written with heavy assistance from a grammar tool or translation software and ended up sounding stiffer or more formal than the rest of a student's annotations. If a flagged comment is genuinely original but reads unnaturally polished next to a student's usual voice, the Humanize tool can adjust the phrasing without changing the actual observation about the reading. This kind of check is a precaution for how writing might be perceived, not a response to anything Perusall itself measures, since the platform has no AI-detection layer of its own to react to.